Extracting the structure behind the content

 

The human has only a single brain to perform a multitude of tasks. Although the issue of a distributed bodily intelligence is an open issue to date, current knowledge offers a vision of human behavior governed by the predominant treatment of cerebral information. Current research advances in artificial intelligence fail to give the intelligent entity capacity for abstraction and plasticity required to solve a set of tasks with a single entity. This makes it necessary to build an artificial intelligence by task, i.e. it should specialize its responses to environmental stimuli to respond optimally to a specific goal.

Using a single physical processing entity to perform a set of tasks forces us to produce and handle powerful mental representations that optimizes the convergence of information. This gives us the ability to construct analogies between tasks, to generalize the existing functions and understand the unknown ones.

By imposing on the artificial intelligence the constraint of usage of common resources to achieve different objectives, we drive it to produce representations of higher degree of abstraction and therefore to deeply understand the environment in which it evolves.

QOPIUS wishes to follow this path and tackle the challenge of A.I. multitasking. QOPIUS draws on research advances in neuroscience to impose this development direction on our artificial intelligence - the Qopius Engine. In fact, the implementation of the principles derived from cognitive sciences and their application in the construction of new algorithms are at the heart of the long-term vision of the company. Specifically, QOPIUS is interested in how knowledge acquired from a particular application can be used to more effectively address other applications with varying degrees of similarities with the first. The knowledge shared to meet various objectives should not simply be the content; it must first be the structure of the task-learning.

QOPIUS therefore wishes to meet several applications needs while focusing on the joint development of the algorithmic structure.